Whisper 1: Full Specifications, Pricing, API Integration, and Use Cases (2026)
What Is Whisper 1?
Whisper 1 is a general-purpose automatic speech recognition (ASR) model from OpenAI. It traces its origins to the Whisper system released on September 21, 2022. Whisper 1 supports multilingual transcription, language identification, and speech-to-English translation. As of July 2026, the hosted API pricing for Whisper 1 is $0.006 per minute of audio.
OpenAI describes Whisper as a speech recognition system built on the Transformer architecture. It is trained using 680,000 hours of multilingual, multitask supervised audio collected from the internet. Its training design supports multilingual speech recognition, language identification, timestamp prediction, and language-to-English translation.
The hosted API model uses the model ID whisper-1. The model takes audio as input and returns text results. It is not a conversational large language model, an inference model, an image model, or a text-to-speech system. OpenAI classifies it as a general-purpose speech recognition model: audio in, text out.
Whisper 1 is commonly used for interview transcription, meeting notes, lecture notes, subtitle creation, media indexing, searchable audio archives, and workflows that translate recorded speech into English.
What Are the Key Specs and Pricing for Whisper 1?
| Spec | Verified Values |
|---|---|
| Provider | OpenAI (as of July 2026) |
| Model family | Whisper (as of July 2026) |
| Model type | General-purpose automatic speech recognition model (as of July 2026) |
| First release date | September 21, 2022 (as of July 2026) |
| Hosted API model ID | whisper-1 (as of July 2026) |
| Gate.AI model ID | openai/stt-whisper-1, see the Gate.AI model card and https://gate.ai/models (as of July 2026) |
| Context window | Official docs do not describe it in the traditional token context-window format (as of July 2026) |
| OpenAI input pricing | $0.006/minute of audio (as of July 2026) |
| Cached input pricing | Not applicable or not stated (as of July 2026) |
| Output pricing | No separate text output price stated; billed by audio duration (as of July 2026) |
| Billing unit | Audio minutes (as of July 2026) |
| Input modality | Audio (as of July 2026) |
| Output modality | Text (as of July 2026) |
| Supported file types | mp3, mp4, mpeg, mpga, m4a, wav, webm (as of July 2026) |
| File upload limit | OpenAI transcription API single file 25 MB (as of July 2026) |
| API access | OpenAI audio API and Gate.AI speech-to-text API (as of July 2026) |
| Translation direction | Supports speech in supported languages to English (as of July 2026) |
| Word-level timestamps | Supported via whisper-1’s timestamp_granularities (as of July 2026) |
| Knowledge cutoff time | Not applicable to the traditional LLM concept; OpenAI does not state it (as of July 2026) |
| Rate limits | Depends on account and usage tier; no single fixed rate specified here (as of July 2026) |
| Fine-tuning support | Hosted whisper-1 API support not confirmed (as of July 2026) |
| Native real-time support | whisper-1 not stated; OpenAI has a separate real-time transcription model and API in its docs (as of July 2026) |
| Batch API support | Docs do not confirm whether whisper-1 supports batch processing (as of July 2026) |
| Speaker separation | Native speaker separation not confirmed (as of July 2026) |
| Tool or function calling | Not applicable to this speech recognition model (as of July 2026) |
| General JSON mode | Not applicable; provides a transcription-specific response format (as of July 2026) |
| Open-source availability | Whisper code and model weights can be obtained separately from the hosted API (as of July 2026) |
| License and usage restrictions | The hosted API follows OpenAI’s service terms; the open-source repo has its own license agreement (as of July 2026) |
OpenAI’s model documentation lists Whisper 1 pricing at $0.006 per minute. Input is audio, and output is text. The model does not publish separate prices for input tokens, cached tokens, or output tokens.
According to Gate.AI’s information in https://gate.ai/models, openai/stt-whisper-1 does not show traditional input/output token pricing. Gate.AI’s API documentation states that whisper-1 is typically billed by audio duration, unlike transcription models billed by tokens.
What Is Whisper 1’s Main Value in Production?
Multilingual Audio Transcription
Whisper 1 converts speech in recorded audio into text in the corresponding language. It fits scenarios like interviews, lectures, user research, internal meetings, podcasts, and media archives. Accuracy depends on factors such as language, accents, microphone quality, background noise, overlapping speakers, and specialized terminology.
Speech-to-English Translation
OpenAI’s official documentation describes two main speech-to-text workflows: (1) transcribing audio in the original language, and (2) directly translating speech in a supported language into English. This can streamline multilingual content moderation by reducing processing steps. However, Whisper 1 does not support direct speech-to-target-language translation into every possible target language.
Word- and Segment-Level Timestamps
Using the timestamp_granularities parameter, you can return timestamps at the segment level, word level, or both. OpenAI explicitly states that this parameter applies to whisper-1. Timestamp data helps with subtitle alignment, transcription navigation, media editing, citation verification, and searchable playback interfaces.
Mainstream Audio and Media Format Support
The Transcriptions API supports common formats like MP3, MP4, MPEG, MPGA, M4A, WAV, and WebM. That reduces the need for format conversion in media pipelines. Still, developers must compress or split large files.
Prompt-Guided Transcription
The API accepts contextual prompt information to help specify expected terms, names, abbreviations, or writing style. Prompts can improve consistency, but they cannot guarantee accurate recognition of rare names or specialized vocabulary.
What Input and Output Modalities Does Whisper 1 Support?
| Modality | Supported? | Notes |
|---|---|---|
| Audio input | Yes | Supports mainstream audio and media file formats |
| Text input | Limited | Optional instructions and transcription parameters; no general text generation |
| Image input | No | Whisper 1 does not process images |
| Video understanding | No | Can process audio content from MP4 files, but does not analyze video frames |
| Text output | Yes | Outputs transcription text or English translation text |
| Audio output | No | Whisper 1 does not generate speech |
| Segment timestamps | Yes | Supports a timestamped response format |
| Word timestamps | Yes | whisper-1 supports timestamp_granularities |
| Native speaker labels | Not confirmed | May require separate speaker separation handling |
| Native real-time dialogue | Not confirmed | For real-time streaming use cases, choose a dedicated real-time transcription model |
OpenAI currently limits a single Transcriptions API upload to 25 MB. It supports mp3, mp4, mpeg, mpga, m4a, wav, and webm. For files exceeding the limit, you must compress or upload them in segments. OpenAI advises against splitting in the middle of sentences, because segment boundaries may lose useful context.
What Are Whisper 1’s Limitations?
Whisper 1 is a dedicated speech recognition model. It cannot independently summarize transcription text, answer questions about recordings, perform topic classification, generate images, synthesize speech, or provide general reasoning. You need other models or an application layer for those tasks.
OpenAI’s 25 MB upload limit adds preprocessing work for long interviews, lectures, meetings, podcasts, and video files. Segmenting can also cause issues like lost context, inconsistent punctuation, duplicated text, or broken sentences.
Its translation workflow outputs only English text. If you need to translate speech directly into a non-English target language, you generally have to set up a separate translation flow.
Whisper 1 does not guarantee accurate speaker attribution. If you require reliable speaker labels, use a verified speaker separation feature or an independent segmentation system, and test on representative recordings.
When there is speaker overlap, the microphone is far away, background music or noise is high, languages switch frequently, names are rare, specialized terms appear, or low-resource languages are used, transcription quality may drop. Context prompts can help with vocabulary recognition, but they cannot be treated as a guarantee of accuracy.
Also, these are limitations of general AI—not unique to Whisper 1. The generated transcription may include reasonable but incorrect words, omissions, or punctuation issues. For transcription in legal, medical, financial, investigative, accessibility, and other safety-critical scenarios, qualified personnel should cross-check against the original audio.
Which Use Cases Fit Whisper 1 Best?
| Use case | Why it works | Key limitations |
|---|---|---|
| Interview transcription | Converts audio into editable, searchable text | Names, quotes, and speaker changes require manual review |
| Meeting notes | Produces text records for easy search and later summaries | Native speaker separation not confirmed |
| Lecture and webinar transcription | Supports long-form speech content, as long as files are prepared properly | Large recordings must be compressed or split |
| Subtitle creation | Outputs timestamped text for easier alignment and editing | Timing and wording still need manual verification |
| Podcast and media indexing | Makes audio content searchable and easier to archive | Music and overlapping speakers can reduce accuracy |
| Multilingual research | Supports original-language transcription | Accuracy depends on language and recording quality |
| Speech-to-English translation | Converts non-English audio into English text | Only supports English as the target language |
| Archival digitization | Converts recordings into searchable text | Poor source audio quality reduces reliability |
| Content moderation assistance | Generates text to help downstream systems moderate | Transcription may lose audio meaning and context |
| Accessibility workflows | Provides draft transcriptions or subtitle sources | Requires human correction before publishing |
A "best use" depends on your specific scenario. Whisper 1 is a good fit when you need a stable, file-based transcription API, English translation, or word-level timestamps. Test on recordings where the target language, accents, acoustic environment, and vocabulary match your real requirements.
How Does Whisper 1 Compare to GPT-4o Transcribe and GPT-4o Mini Transcribe?
| Comparison dimension | Whisper 1 | GPT-4o Transcribe | GPT-4o Mini Transcribe | Use cases |
|---|---|---|---|---|
| Primary purpose | General multi-language speech recognition | OpenAI’s next-generation transcription model | OpenAI’s next-generation smaller transcription model | All three work for speech-to-text workflows |
| Audio input | Supported | Supported | Supported | For transcription use cases |
| Text output | Supported | Supported | Supported | All three output text transcription results |
| OpenAI pricing | $0.006/minute audio | Check the latest OpenAI pricing before deployment | Check the latest OpenAI pricing before deployment | Recommend comparing total cost using real-time pricing |
| English translation | Whisper 1 clearly supports it | Please confirm whether this endpoint currently supports it | Please confirm whether this endpoint currently supports it | Whisper 1 is for workflows that already do speech-to-English |
| Word-level timestamps | Whisper 1 clearly supports it | Please confirm whether this feature is supported | Please confirm whether this feature is supported | Whisper 1 fits applications already producing timestamp-based outputs |
| API maturity | Hosted model, long integration history | Next-generation model | Next-generation efficient model | For existing systems, prioritize compatibility; for new apps, run comparison tests |
| Native speaker separation | Not confirmed | Please check the latest model docs | Please check the latest model docs | If you need speaker labels, verify the feature |
| Real-time applications | Native real-time support not stated | Please check the latest real-time support | Please check the latest real-time support | For real-time scenarios, use a dedicated real-time transcription model |
No transcription model fits every scenario. Whisper 1 is best suited for existing integrations that rely on its model ID, translation behavior, or timestamp formats. For new applications, test Whisper 1 and next-generation transcription models on representative audio, and compare accuracy, latency, response characteristics, and overall cost.
How Do You Access Whisper 1 via Gate.AI?
According to Gate.AI’s model card (https://gate.ai/models), Gate.AI’s model ID is:
openai/stt-whisper-1
Gate.AI provides a OpenAI-compatible API base URL:
XX/openai/v1
Its speech-to-text interface documentation includes: POST /audio/transcriptions, Bearer-token authentication, a multipart/form-data request body, synchronous text output, information usage, and billing by duration.
Python Example
Python import os
from pathlib import Path
import OpenAI from openai
audio_path = Path("audio.mp3")
if audio_path.is_file():
raise FileNotFoundError(f"Audio file not found: {audio_path}")
client = OpenAI(
api_key=os.environ["GATEAI_API_KEY"],
base_url="XX/openai/v1",
)
with audio_path.open("rb") as audio_file:
transcription = client.audio.transcriptions.create(
model="openai/stt-whisper-1",
file=audio_file,
)
print(transcription.text )
curl Example
``` Bash curl —request POST \
—url "XX/openai/v1/audio/transcriptions" \
—header "Authorization: Bearer ${GATEAI_API_KEY}" \
—form "file=@audio.mp3" \
—form "model=openai/stt-whisper-1" ```
The example above stores the API key in the environment variable GATEAI_API_KEY to avoid embedding it directly in source code. Before deploying an application, you should also validate file type, file size, error responses, retry mechanisms, and account quotas.
Gate.AI describes the speech-to-text interface as a synchronous one. Therefore, for long-duration or high-concurrency tasks, it’s recommended to design sensible request timeouts, queues, file validation, and retry strategies.
Developers can also access the model directly through OpenAI’s /v1/audio/transcriptions and /v1/audio/translations APIs by using OpenAI’s model ID whisper-1.
Frequently Asked Questions
What is Whisper 1’s context window size?
OpenAI does not define Whisper 1’s context window using a traditional token context window, because it is a speech transcription model rather than a chat-style LLM. The more relevant API constraint is the 25 MB file upload limit of the OpenAI transcription API.
What is Whisper 1’s pricing?
OpenAI prices Whisper 1 at $0.006 per minute of audio (as of July 2026). No separate prices for input tokens, cached tokens, or output tokens have been published. Gate.AI records whisper-1 as billed by duration; check your account’s current pricing before deployment.
How do developers access Whisper 1?
Developers can access Whisper 1 directly via OpenAI’s audio API using the model ID whisper-1. According to Gate.AI’s model card, you can also use Gate.AI’s OpenAI-compatible /audio/transcriptions endpoint and call openai/stt-whisper-1.
What scenarios is Whisper 1 suitable for?
Whisper 1 is suitable for interview recordings, meetings, lectures, subtitles, podcasts, multilingual transcription, searchable audio archives, and speech-to-English translation. Accuracy depends on language, accents, recording quality, overlapping speakers, background noise, and specialized vocabulary. For important transcriptions, you should do manual review.


